Strip away the chat interface and a large language model does exactly one thing: it predicts the next token. Everything else — answering questions, writing code, holding a conversation — is that single ability, applied over and over. A token is a chunk of text, usually a short word or a fragment of one. The model reads the tokens so far and produces a probability for every possible next token, then one is chosen. Append it, and repeat. The startling discovery of the last decade is how much competence falls out of this humble objective once the model is large enough and has read enough text. Fluency, translation, a surprising amount of reasoning — none of it was programmed in. It emerged from prediction at scale.

Byte: why 'strawberry' has three r's but the model miscounts

Models don't see letters — they see tokens. 'Strawberry' might arrive as two or three tokens, none of which is a single 'r'. Asking a model to count letters is asking it to reason about something it never directly perceives. It's not stupidity; it's a consequence of tokenization. Newer models are trained to compensate, but the lesson holds: the model's world is made of tokens, not characters.

1. 1.What is the single core task a language model is trained to perform?

   Store and retrieve facts from a databasePredict the next token given the tokens so farTranslate every input into English firstSearch the internet for the best answer

2. 2.A 'token' is best described as:

   A single letter of the alphabetOne complete sentenceA chunk of text — often a short word or word-fragmentA unit of payment only, with no role in processing